收斂 NemoTron 舊回覆行動

config.py

@@ -320,7 +320,7 @@ YOUTUBE_API_KEY = os.getenv('YOUTUBE_API_KEY', '')
 # ==========================================
 # 系統版本與路徑
 # ==========================================
-SYSTEM_VERSION = "V10.269"
+SYSTEM_VERSION = "V10.270"
 LOG_FILE_PATH = os.path.join(BASE_DIR, 'logs/system.log')
 public_url = PUBLIC_URL  # 用於模板顯示
 

docs/AI_INTELLIGENCE_MODULE_SOT.md

@@ -2,7 +2,7 @@
 
 > **最後更新**: 2026-05-19 (台北時間)
 > **狀態**: 🟢 四 AI Agent 自動化閉環已落地；LLM 路由紅線升級為 Ollama-first 三主機級聯，Gemini 僅備援 / 鎖定場景
-> **適用版本**: V10.269
+> **適用版本**: V10.270
 
 ---
 
@@ -107,7 +107,7 @@ SQL漏斗(~300筆)
 - ElephantAlpha 使用 NVIDIA NIM hosted API；production 預設模型為 `nvidia/llama-3.3-nemotron-super-49b-v1.5`，`ELEPHANT_ALPHA_FALLBACK_MODELS` 需保留至少一個可呼叫備援；403/404、408/409/425/429、5xx、timeout 與 connection error 必須嘗試下一個模型。
 - ElephantAlpha L3 HITL 只允許發送有實證、可審核、可行動的升級告警；價格類 trigger 無 Hermes 具體威脅時，只記錄 suppressed escalation telemetry 與 cooldown，不寫 pending `human_review`，不發 Telegram 空告警。
 - `resource_optimization` 不再交給 LLM 生成「預期效益 / 已執行」敘事。此 trigger 必須先由程式量測 `action_plans` backlog、P1/P2 數、pending_review、逾時項目與 CPU load；只有 CPU 達門檻、P1/P2 積壓或逾時積壓才發 Telegram「資源壓力告警」。單純 queue 大但 CPU 正常只記錄 telemetry，不派發 Hermes/NemoTron、不宣稱 48 小時效益。
-- `resource_optimization` 會先執行 `ActionPlanHygieneService` 清理過期噪音：只關閉超過 72 小時的 `code_review_fix` / `openclaw_recommendation` 類 advisory action_plans，將狀態改為 `auto_disabled` 或 `rejected` 並寫入 `metadata_json.hygiene_history`；不刪資料、不碰 NemoTron 業務行動。
+- `resource_optimization` 會先執行 `ActionPlanHygieneService` 清理過期噪音：只關閉超過 72 小時的 `code_review_fix` / `openclaw_recommendation` 類 advisory action_plans，以及 NemoTron `direct_response/reply_simple` 舊聊天回覆計畫；將狀態改為 `auto_disabled` 或 `rejected` 並寫入 `metadata_json.hygiene_history`。不刪資料，也不碰 NemoTron human_review / pricing / tool action 類業務行動。
 - OpenClaw/Hermes embedding 優先呼叫 Ollama `/api/embed`，只在舊節點不支援時 fallback `/api/embeddings`；timeout 由 `EMBEDDING_TIMEOUT` / `OLLAMA_EMBED_TIMEOUT` 控制。
 - PPT 自動產線由 `momo-scheduler` 依節奏執行 `run_ppt_auto_generation_task(schedule_kind)`：每日 20:30 產日報、週一 20:40 產週報/市場情報、每月 1 日 20:50 產月報與管理型簡報、季初 21:00 產季報、半年初 21:10 產半年報、年初 21:20 產年報，再交給 22:00 `ppt_vision_audit` 做視覺審核；每次嘗試會寫入 `ppt_generation_runs`，`/observability/ppt_audit_history` 以精準參數檢查目標版本是否已產生，並可用 `/observability/ppt_audit/generate_missing` 手動補齊缺漏，總開關為 `PPT_AUTO_GENERATION_ENABLED`。PPT vision 需 `PPT_VISION_ENABLED=true` 與容器內 LibreOffice；`/observability/ppt_audit_file/<filename>` 會把 PPTX 轉成 PDF 快取供站內線上預覽，原始 PPTX 仍保留下載。
 

services/action_plan_hygiene.py

@@ -24,6 +24,7 @@ CLOSABLE_STATUSES = frozenset({"pending", "auto_pending", "pending_review"})
 SOURCE_TARGET_STATUS = {
     "code_review_fix": "auto_disabled",
     "code_review_pipeline": "auto_disabled",
+    "nemotron_direct_response": "auto_disabled",
     "openclaw_recommendation": "rejected",
     "openclaw": "rejected",
 }
@@ -51,15 +52,6 @@ def _coerce_datetime(value: Any) -> Optional[datetime]:
     return None
 
 
-def _source_for_row(row: Any) -> str:
-    return str(
-        _row_get(row, "action_type")
-        or _row_get(row, "plan_type")
-        or _row_get(row, "created_by")
-        or "unknown"
-    )
-
-
 def _parse_metadata(raw: Any) -> Dict[str, Any]:
     if isinstance(raw, dict):
         return dict(raw)
@@ -72,6 +64,39 @@ def _parse_metadata(raw: Any) -> Dict[str, Any]:
         return {"legacy_metadata_raw": str(raw)[:1000]}
 
 
+def _parse_payload(raw: Any) -> Dict[str, Any]:
+    if isinstance(raw, dict):
+        return dict(raw)
+    if not raw:
+        return {}
+    try:
+        parsed = json.loads(str(raw))
+        return parsed if isinstance(parsed, dict) else {}
+    except Exception:
+        return {}
+
+
+def _source_for_row(row: Any) -> str:
+    created_by = str(_row_get(row, "created_by") or "")
+    if created_by == "nemotron":
+        payload = _parse_payload(_row_get(row, "payload"))
+        actions = payload.get("action_plan") if isinstance(payload.get("action_plan"), list) else []
+        is_direct_response = payload.get("dispatch_to") == "direct_response"
+        is_reply_only = bool(actions) and all(
+            isinstance(action, dict) and action.get("action") == "reply_simple"
+            for action in actions
+        )
+        if is_direct_response or is_reply_only:
+            return "nemotron_direct_response"
+
+    return str(
+        _row_get(row, "action_type")
+        or _row_get(row, "plan_type")
+        or created_by
+        or "unknown"
+    )
+
+
 def build_action_plan_hygiene_preview(
     rows: Iterable[Any],
     *,
@@ -140,12 +165,12 @@ class ActionPlanHygieneService:
         try:
             rows = session.execute(text("""
                 SELECT id, status, priority, created_at, action_type, plan_type,
-                       created_by, description, metadata_json
+                       created_by, description, metadata_json, payload
                 FROM action_plans
                 WHERE status IN ('pending', 'auto_pending', 'pending_review')
                   AND (
                     action_type IN ('code_review_fix', 'openclaw_recommendation')
-                    OR created_by IN ('code_review_pipeline', 'openclaw')
+                    OR created_by IN ('code_review_pipeline', 'openclaw', 'nemotron')
                   )
                 ORDER BY created_at ASC
             """)).fetchall()
@@ -163,12 +188,12 @@ class ActionPlanHygieneService:
         try:
             rows = session.execute(text("""
                 SELECT id, status, priority, created_at, action_type, plan_type,
-                       created_by, description, metadata_json
+                       created_by, description, metadata_json, payload
                 FROM action_plans
                 WHERE status IN ('pending', 'auto_pending', 'pending_review')
                   AND (
                     action_type IN ('code_review_fix', 'openclaw_recommendation')
-                    OR created_by IN ('code_review_pipeline', 'openclaw')
+                    OR created_by IN ('code_review_pipeline', 'openclaw', 'nemotron')
                   )
                 ORDER BY created_at ASC
             """)).fetchall()

tests/test_action_plan_hygiene.py

@@ -38,12 +38,40 @@ def test_action_plan_hygiene_preview_closes_only_stale_advisory_sources():
             "action_type": "openclaw_recommendation",
             "description": "fresh advisory",
         },
+        {
+            "id": 5,
+            "status": "pending",
+            "priority": 3,
+            "created_at": now - timedelta(hours=110),
+            "created_by": "nemotron",
+            "payload": {
+                "dispatch_to": "direct_response",
+                "action_plan": [{"action": "reply_simple"}],
+            },
+            "description": "old chat response",
+        },
+        {
+            "id": 6,
+            "status": "pending",
+            "priority": 3,
+            "created_at": now - timedelta(hours=110),
+            "created_by": "nemotron",
+            "payload": {
+                "dispatch_to": "human_review",
+                "action_plan": [{"action": "flag_for_human_review"}],
+            },
+            "description": "must stay",
+        },
     ]
 
     preview = build_action_plan_hygiene_preview(rows, now=now, stale_hours=72)
 
-    assert preview["candidate_count"] == 2
-    assert preview["by_source"] == {"openclaw_recommendation": 1, "code_review_fix": 1}
-    assert {item["id"] for item in preview["candidates"]} == {1, 2}
+    assert preview["candidate_count"] == 3
+    assert preview["by_source"] == {
+        "openclaw_recommendation": 1,
+        "code_review_fix": 1,
+        "nemotron_direct_response": 1,
+    }
+    assert {item["id"] for item in preview["candidates"]} == {1, 2, 5}
     target_by_id = {item["id"]: item["to_status"] for item in preview["candidates"]}
-    assert target_by_id == {1: "rejected", 2: "auto_disabled"}
+    assert target_by_id == {1: "rejected", 2: "auto_disabled", 5: "auto_disabled"}